Classical and Bayesian approaches to reconstructing genetic regulatory networks with hidden facto

We have used state space models to reverse engineer transcriptional networks from highly replicated gene expression profiling time series data obtained from a well-established model of T cell activation. State space models are a class of dynamic Bayesian networks which assume that the observed measurements depend on some hidden state variables which evolve according to Markovian dynamics. These hidden variables can capture effects which cannot be measured in a gene expression profiling experiment, for example: genes that have not be included in the microarray, levels of regulatory proteins, the effects of mRNA and protein degradation, etc. We have approached the problem of inferring the model structure of these state-space models using both classical and Bayesian methods. A bootstrap procedure is used to derive classical confidence intervals for parameters representing `gene-gene' interactions over time. Variational approximations are used to perform the analogous model selection task in the Bayesian context.

Certain interactions are robustly reproduced in both the classical and the Bayesian analysis of these regulatory networks. The resulting models reflect many of the dynamics of an activated T cell. In particular, they reveal the integrated activation of cytokines, proliferation, and adhesion following activation and place JunB and JunD at the centre of the mechanisms that control apoptosis and proliferation. These mechanisms are key for clonal expansion and to control the long term behavior (e.g. programmed cell death) of these cells. Our models provide a methodology for the development of rational and experimentally testable hypotheses concerning the dynamics of T cell activation.

We develop the standard SSMM by incorporating multiple alignment profiles into the model to achieve comparable results to other contemporary methods. We propose a novel parameterized model as the likelihood function for the SSMM to exploit the evolutionary information provided by the profiles. The parametric model is composed of two components: one comes from a Dirichlet distribution with free parameters to capture the segmental dependency; another is a position-specific distribution to capture the helical capping signals. Moreover, we incorporate the long-range interaction information in beta-sheets into the modelling. We describe a Markov Chain Monte Carlo sampling scheme to perform inference in this model, and then demonstrate the capability of the parametric SSMM to carry out inference on beta-sheet contact maps in the Bayesian segmental framework. This ability to infer contact maps represents one of the advantages of a probabilistic modelling approach over the traditional discriminative approach to protein secondary structure prediction.

We have used state space models to reverse engineer transcriptional networks from highly replicated gene expression profiling time series data obtained from a well-established model of T cell activation. Our models represent the dynamics of T cell activation and provide a methodology for the development of experimentally testable hypotheses.